A Process-Aware Comparison of Mountain Precipitation Estimates in The East River Watershed

The quality of precipitation and temperature data limit our understanding of regional and watershed scale hydrologic processes, especially in mountainous regions. Complex terrain induces spatially variable rates of orographic precipitation enhancement. This spatial variability is difficult to capture with existing observational networks, since precipitation gauges are sparse and possess uncertainties of their own (undercatch, for example). Geostatistical datasets use different assumptions and data sources, but in general apply weighted interpolation functions to create continuous precipitation fields from sparse observations. Increasingly, high resolution Numerical Weather Prediction (NWP) models have proven to be capable alternatives that reasonably simulate precipitation in mountain watersheds. In this study, we characterize the differences in precipitation and temperature between a 1km configuration of the Weather Research and Forecasting (WRF) model and commonly used geostatistical datasets including PRISM, DayMet, and NLDAS across Colorado’s Upper Gunnison watershed. We apply a supervised clustering technique to classify the orographic precipitation ‘fingerprints’, including factors such as wind speed, wind direction, integrated vapor transport, and atmospheric stability, and interpret the discrepancies between the data products in the context of these clusters. We find that the WRF model has the highest orographic precipitation gradient (OPG), expressed as fraction of precipitation at elevation relative to the valley-bottom, and that NLDAS has a particularly low bias relative to the other datasets. Differences between the various products are on the order of 20 cm per year or precipitation per year, or approximately 20-25% of the total. These results highlight the need for concerted, field based campaigns to further validate modeled mountain hydrometeorological processes.